Targeting iron-associated protein Ftl1 in the brain of old mice improves age-related cognitive impairment

Animal models

The following mouse lines were used: C57BL/6J mice (The Jackson Laboratory, line 000664), C57BL/6 aged mice (National Institutes of Aging) and B6;129-Gt(ROSA)26Sor^{tm1(CAG-cas9*,-EGFP)Fezh}/J (The Jackson Laboratory, line 024857). All studies were done in young (2–3 months) or aged (18–22 months) male mice that were not involved in any previous procedures. The numbers of mice used to result in statistically significant differences were calculated using standard power calculations with α = 0.05 and a power of 0.8. We used an online tool (https://www.stat.uiowa.edu/~rlenth/Power/index.html) to calculate power and sample size based on experience with the respective tests, variability of the assays and inter-individual differences within groups. Mice were housed under specific pathogen-free conditions under a 12-hour light/dark cycle, with humidity maintained at 30–70% and temperature at 68–79 °F (20–26 °C). All animal handling and use were in accordance with institutional guidelines approved by the University of California, San Francisco (UCSF) Institutional Animal Care and Use Committee (IACUC).

Tissue collection

Mice were anesthetized with 87.5 mg kg⁻¹ ketamine and 12.5 mg kg⁻¹ xylazine and transcardially perfused with ice-cold PBS. Tissues were removed and processed for subsequent analysis. To process the brains, either the hippocampus was subdissected and snap frozen or the whole brain was fixed in phosphate-buffered 4% paraformaldehyde (pH 7.4) at 4 °C for 48 hours before cryoprotection with 30% sucrose.

Primary neuron cultures

Hippocampal and cortical neurons were isolated from E17 C57Bl/6J mouse embryos using the papain dissociation system (Worthington, cat. no. LK003153). Cells were plated at 100,000 cells per well on 12-mm poly-l-lysine-coated glass coverslips (Carolina, cat. no. 633009) in 24-well plates. Cultures were maintained at 37 °C and 5% CO₂ in neurobasal medium (Thermo Fisher Scientific, cat. no. 21103049) supplemented with B-27 (cat. no. 17504044), GlutaMAX (cat. no. 35050061) and penicillin–streptomycin (cat. no. 15140122). Media were partially changed every 4–5 days. On day in vitro 8 (DIV8), neurons were infected with lentivirus at multiplicity of infection (MOI) = 1 and processed 10 days later (DIV18) for immunocytochemistry. Cells were fixed with 4% paraformaldehyde (10 minutes), washed and stained with MAP2 (Sigma-Aldrich, cat. no. M1406, RRID: AB_477171), anti-GFP (Aves Labs, cat. no. GFP-1020) or anti-turboGFP (Thermo Fisher Scientific, cat. no. PA5–22688, RRID: AB_2540616) antibodies to confirm infection, along with Hoechst 33342 (Thermo Fisher Scientific, cat. no. H3570) for nuclear staining.

Viral plasmids and viruses

Lentiviral constructs for Ftl1 overexpression, knockdown and conditional knockout were generated using standard molecular cloning techniques. For overexpression, murine Ftl1 coding sequence (with partial untranslated regions) was polymerase chain reaction (PCR) amplified from adult mouse hippocampal cDNA and cloned into pENTR-D-TOPO (Thermo Fisher Scientific, cat. no. K240020), sequence verified and subcloned into a synapsin–IRES–eGFP lentiviral backbone using NheI and EcoRI. The synapsin–IRES–eGFP plasmid alone served as control. For knockdown, shRNAs targeting Ftl1 (SH1–SH4) were cloned into the pGreenPuro system (System Biosciences), with a luciferase-targeting shRNA as control^15,42. For CRISPR-based knockout vectors, six gRNAs targeting Ftl1 exons 2–4 were designed using CHOPCHOP⁴³ and cloned into LentiCRISPR v2 (ref. ⁴⁴). gRNA sequences included CGTATTTGTAACCTCCCGCA and CGCAGACTGGCGCGCCCCAG, among others. A non-targeting gRNA (GGCGAGGGCGATGCCACCTA) served as control. Validated gRNA sequences were subcloned into a Syn1-Cre lentiviral vector (Addgene, no. 68841; ref. ⁴⁵) using PacI and NheI/XbaI. All constructs were verified by Sanger sequencing. Lentivirus production followed previously described protocols⁴⁶. Ftl1 expression changes were confirmed by qRT–PCR and western blot in mouse cells.

Lentivirus production

HEK293T cells (American Type Culture Collection, cat. no. CRL-11268) were transfected using a 4:3:1-µg ratio of lentiviral construct, psPAX2 (Addgene, no. 12260, gift from D. Trono) and pCMV-VSV-G (Addgene, no. 8454, RRID: Addgene_8454)⁴⁷. After 48 hours, supernatants were cleared by centrifugation (5 minutes, 1,000g) and filtration (0.45 μm) and then concentrated by ultracentrifugation (24,000 r.p.m, 1.5 hours). Pellets were resuspended in PBS. Primary neurons were infected at MOI = 1. For in vivo injections, viral solutions were diluted to 1.0 × 10⁸ viral particles per milliliter (vp/ml).

Confocal imaging and quantification

For hippocampal synaptic proteins, confocal images were acquired on an LSM 900 microscope (Zeiss) with a ×63 objective as z stacks (16 optical sections, 0.33-µm step size) per sample. Maximum intensity projections were generated for individual puncta analysis. Synapsin (555 nm) and gephyrin or PSD95 (647 nm) were quantified using the Analysis module. For co-localization, z stacks were split into four approximately 1-µm sub-stacks, each projected to two dimensions. Co-localization was assessed using the ZEN Toolkit 3D. For primary cell cultures, confocal z stacks (12 sections, 0.80-µm step) were acquired on an LSM 900 microscope (Zeiss) using a ×20 objective. Neurite length was quantified with NeuronJ (version 1.4.3) and imagescience.jar in FIJI (ImageJ version 1.54f). Sholl analysis was performed using the Simple Neurite Tracer in the Neuroanatomy plugin. Concentric rings (10-µm spacing, up to 300 µm) centered on the soma were used to quantify dendrite complexity by counting neurite intersections, which were summed to calculate total arborization and plotted as a function of distance.

Real-time ATP rate using Seahorse

Mitochondrial ATP production was assessed in primary neurons (75,000 cells per well) using the Seahorse XFe24 Analyzer (Agilent Technologies). Cells were plated on Seahorse 24-well plates pre-coated with 50 μl of poly-l-lysine (10 µg ml⁻¹; Millipore, cat. no. A-005-C) after Ftl1 overexpression or knockdown. On the day of the assay, cells were equilibrated for 1 hour in Seahorse assay medium (DMEM supplemented with 10 mM glucose, 1 mM pyruvate and 2 mM glutamine; Agilent Technologies). ATP production was measured using sequential injections of 1.5 µM oligomycin and 0.5 µM rotenone/antimycin (Agilent Technologies, cat. no. 103592-100).

Stereotaxic injections

Procedures were adapted from Lin et al.⁴⁶. Mice were anesthetized with 2% isoflurane in oxygen (2 l min⁻¹) and secured in a stereotaxic apparatus. Ophthalmic ointment was applied to prevent corneal drying, and fur over the skull was trimmed. Adeno-associated virus solutions (1.0 × 10⁸ vp/ml) were bilaterally injected into the hippocampal CA1 and dentate gyrus at the following coordinates (from bregma: AP: −2.0 mm, ML: ±1.5 mm; from skull surface: DV: −1.8 mm and −2.1 mm) using a 5-μl 26s-gauge Hamilton syringe. A volume of 2 μl per site was infused over 10 minutes (0.2 μl min⁻¹). To minimize backflow, the needle was left in place for 8 minutes after injection and then partially retracted and held for another 2 minutes. Incisions were closed with silk sutures and VetBond. Mice received subcutaneous saline, enrofloxacin, carprofen and buprenorphine postoperatively and were monitored during recovery.

Neuronal nuclei isolation

Neuronal nuclei were isolated from flash-frozen hippocampi following a modified version of Krishnaswami et al.⁴⁸. Tissue was homogenized in 750 μl of ice-cold homogenization buffer (250 mM sucrose, 25 mM KCl, 5 mM MgCl₂, 10 mM Tris (pH 8.0), 0.1% Triton X-100, 1 μM DTT, RNase and protease inhibitors) using a dounce homogenizer (Wheaton, cat. no. 357538; 12 loose and 20 tight pestle strokes). Homogenates were passed through a 40-μm filter and centrifuged (500 relative centrifugal force (RCF), 6 minutes, 4 °C) and then washed and pelleted again under the same conditions. Pellets were resuspended in staining buffer (0.5% BSA in PBS) and incubated on ice for 15 minutes. Alexa Fluor 488-conjugated anti-NeuN antibody (Millipore, cat. no. MAB377X, RRID: AB_2149209) was added at 1:250, and samples were rotated at 4 °C for 1 hour. After two washes in staining buffer, Hoechst 33342 was added (0.01 μg ml⁻¹), and samples were filtered through 35-μm mesh FACS tubes prior to sorting.

Nuclei isolation for single-nucleus RNA-seq

Neuronal nuclei were isolated from flash-frozen hippocampi using a modified version of the 10x Genomics protocol. Tissue was homogenized in 500 μl of NP-40 lysis buffer (10 mM Tris-HCl (pH 7.4), 10 mM NaCl, 3 mM MgCl₂, 0.1% NP-40 substitute, 1 mM DTT and RNase inhibitor) with 20 loose and 25 tight pestle strokes (Wheaton, cat. no. 357538), followed by the addition of another 500 μl of buffer and 7-minute incubation on ice. Homogenates were filtered (40 μm) and centrifuged (500 RCF, 5 minutes, 4 °C) and then washed in PBS with 1% BSA and RNase inhibitor. After another spin, samples were resuspended in 400 μl of wash buffer containing Hoechst 33342 (1:10,000), incubated for 5 minutes on ice, filtered through 35-μm mesh and sorted using a BD FACSAria II.

FACS

Nuclei were sorted on a BD FACSAria Fusion with a 70-μm nozzle and a flow rate of 1–2.5 ml min⁻¹. Nuclei were first gated by forward and side scatter and then gated for doublets with height and width. Nuclei that were both Hoechst⁺ and NeuN⁺ (neuronal nuclei isolation) or only Hoechst⁺ (single-nucleus isolation) were sorted into TRI Reagent (Sigma-Aldrich, T9424) for RNA analysis.

Library generation and RNA-seq

RNA-seq libraries were generated using a modified Smart-seq2 protocol (Picelli et al.⁴⁹). After RNA isolation with TRI Reagent, 8 ng of high-quality RNA was reverse transcribed using SuperScript II (Thermo Fisher Scientific, cat. no. 18064-014) with anchored poly-dT and TSO primers. cDNA was amplified (10 PCR cycles) using KAPA HiFi HotStart polymerase (cat. no. KK2601) and purified with AMPure XP beads (Beckman Coulter, cat. no. A63881). Quality was assessed by Qubit fluorometry. For library construction, 2 ng of cDNA was fragmented (approximately 500 bp) with Nextera Tn5 transposase (Illumina, cat. no. FC-131-1096) and PCR amplified (12 cycles) with indexed Nextera primers (Illumina, cat. no. FC-131-1002). Final libraries were bead purified, analyzed with an Agilent Bioanalyzer and sequenced on an Illumina NovaSeq (paired-end 2 × 150 bp, SP flow cell).

RNA-seq analysis

Reads were aligned to the mm10 mouse transcriptome using STAR version 2.7.3a⁵⁰ with ENCODE-recommended settings. Gene-level quantification was performed with RSEM version 1.3.1 (ref. ⁵¹), and differential expression analysis was carried out in R version 4.0.2 using DESeq2 version 1.28.1 (ref. ⁵²). Full pipeline details (version 2.1.2) are available at https://github.com/emc2cube/Bioinformatics/. Gene Ontology enrichment was performed using Panther.

Single-nuclei RNA-seq analysis

Nuclei were submitted to the UCSF IHG Genomics Core for processing with the 10x Genomics Chromium Single Cell 3′ kit. Approximately 10,000 nuclei per sample were captured, and cDNA libraries were prepared following manufacturer protocols (10x Genomics). Libraries were sequenced on an Illumina NovaSeq 6000 S2. Base calls were demultiplexed using Cell Ranger version 7.1, and raw FASTQ files were processed to generate expression matrices, including intronic reads. Each sample yielded approximately 13,100–15,950 nuclei, with a mean depth of 27,000 reads per nucleus and approximately 44% sequencing saturation. Data were analyzed in R version 4.2.2. Ambient RNA contamination was removed using SoupX, and samples were integrated with Seurat (2,000 variable genes, 20 dimensions). Nuclei expressing fewer than 300 or more than 5,000 genes, or more than 2% mitochondrial transcripts, were excluded. Genes detected in fewer than three nuclei were filtered out. Doublets and technical artifacts were removed, and expression was log normalized. Dimensionality reduction was performed using PCA, and clustering was conducted with the Louvain algorithm on the first 20 principal components. UMAP was applied for two-dimensional visualization. Differential expression was calculated using Seurat (min.pct = 0.05, log fold change > 0.15, pseudocount = 0.1). Genes inconsistent across replicates were excluded. Visualizations included violin plots, UMAPs and heatmaps from average expression matrices. Volcano plots were generated using EnhancedVolcano.

Synaptosomes protein sample preparation

To obtain sufficient synaptosomes for proteomic analysis, hippocampus and cortex from multiple mice were pooled. Synaptosomes were isolated following a modified protocol from Trinidad et al.⁵³. Tissue was homogenized in sucrose buffer with phosphatase inhibitors (Na₃VO₄, NaF, Na₂MoO₄, sodium tartrate, fenvalerate and okadaic acid) and PUGNAc, followed by differential centrifugation. Synaptic membranes were collected at the 1.0–1.2 M sucrose interface and pelleted. Frozen pellets were resuspended in 50 mM ammonium bicarbonate with 6 M guanidine hydrochloride, phosphatase inhibitor cocktails II/III and PUGNAc. Protein concentration was measured by BCA assay (Thermo Fisher Scientific). Each sample (1.5 mg) was reduced with 2.5 mM TCEP at 56 °C for 1 hour, alkylated with 5 mM iodoacetamide (45 minutes, room temperature, dark), diluted to 1 M guanidine and digested overnight at 37 °C with sequencing-grade trypsin (1:50 enzyme:substrate). Peptides were acidified with formic acid, desalted (C18 Sep-Pak) and dried by SpeedVac. Tryptic peptides were labeled with TMT-6plex reagents (Thermo Fisher Scientific) according to the manufacturer’s protocol: TMT126–131 for three young and three aged samples. Labeling efficiency was confirmed on a Q Exactive Plus Orbitrap. Labeled peptides were quenched with 5% hydroxylamine, pooled, desalted and dried. Peptides were separated using high-pH reverse-phase chromatography on a Gemini 5-μm C18 column (4.6 × 150 mm; Phenomenex). A 9–49% acetonitrile gradient in 20 mM ammonium formate (pH 10) was run over 20 ml at 550 μl min⁻¹, collecting 60 fractions. Fractions were combined (two per sample) into 12 final samples and dried for downstream analysis.

Mass spectrometry analysis

Tandem mass tag (TMT)-labeled peptides were analyzed on an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) coupled to a NanoAcquity UPLC system (Waters). Peptides were separated on a 50-cm × 75-μm ID, 2-μm C18 EASY-Spray column using a linear gradient from 3.5% to 30% solvent B over 185 minutes. MS1 scans were acquired from 375 m/z to 1,500 m/z at 120,000 resolution (AGC target: 4.0 × 10⁵). Ions with charge states 2–7 were selected using a 1.0-m/z window, with dynamic exclusion (30 seconds) and MIPS filtering enabled. Higher-energy collisional dissociation MS2 spectra were acquired using stepped collision energies (30%/35%/40%) and detected in the Orbitrap at 30,000 resolution (AGC: 5.0 × 10⁴; maximum injection time: 100 ms). The scan cycle was set to 3 seconds. Peaklists were generated using Proteome Discoverer version 2.2 and searched against the SwissProt Mus musculus database (downloaded 6 September 2016, with decoy sequences) using Protein Prospector version 5.21.1. Searches assumed trypsin specificity with up to two missed cleavages. Fixed modifications included carbamidomethylation (Cys) and TMT-6plex labels (Lys and N termini); variable modifications included N-terminal acetylation, oxidation (Met), pyro-glutamate formation (Gln) and N-terminal Met loss (with or without acetylation). A maximum of two variable modifications per peptide was allowed. Identifications were filtered to a 1% false discovery rate at both peptide and protein levels.

TMT mass spectrometry data analysis

Data were filtered to only include peptides unique to a single protein. Quantitation of TMT data was performed by calculating ratios of reporter ion peak intensities between conditions along with variance for each ratio and median normalized. Protein abundances were normalized by the median of ratio distributions. The age-dependent changes were determined using a normalized median log₂ Aged/Young ratio of at least 1.0, corresponding to a 2.0 fold change with age and a −log₁₀ P value greater than 1.3.

RT–qPCR

To quantify mRNA expression levels, equal amounts of cDNA were synthesized using a High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, 4368813) and then mixed with SYBR Fast mix (Kapa Biosystems) and primers. β-actin was amplified as an internal control. RT–qPCR was performed in the CFX384 Real Time System (Bio-Rad). Each sample and primer set was run in triplicate, and relative expression levels were calculated using the 2^−ΔΔCt method.

Western blot analysis

Western blotting was performed on both mouse hippocampal tissue and primary hippocampal neurons. For tissue studies, hippocampi were dissected, flash frozen and pulverized prior to lysis in RIPA buffer (Abcam, ab156034) with protease inhibitors (Sigma-Aldrich, cat. no. 4693116001). Lysates were mixed with 4× LDS sample buffer (Invitrogen, NP0008), resolved by SDS-PAGE (Invitrogen) and transferred to nitrocellulose membranes. Membranes were blocked in 5% milk in TBST and probed with the following primary antibodies: anti-GAPDH (Abcam, ab8245, 1:5,000), β-tubulin (Covance, MMS-435P, 1:1,000), NR2A (Sigma-Aldrich, 07-632, 1:1,000), synapsin (Abcam, ab18814, 1:1,000) and AMPA receptor (Abcam, ab109450, 1:3,000). HRP-conjugated donkey anti-mouse (Invitrogen, A15999, 1:2,000) and anti-rabbit (GE Healthcare, NA934V, 1:2,000) secondaries were used for detection with ECL reagents (GE Healthcare). Signals were imaged with a ChemiDoc system (Bio-Rad) and quantified in FIJI/ImageJ (version 2.0.0).

For cell-based experiments, primary neurons were cultured from E17 mouse embryos, treated as indicated, lysed under the same conditions and processed in parallel. Quantification was normalized to loading controls (GAPDH, β-tubulin or β-actin) and to the control group. Each tissue datapoint represents an individual animal; each cell culture datapoint represents a biological replicate. Statistical comparisons were made using two-tailed unpaired Student’s t-tests.

Iron quantification

Iron imaging was performed using a modified protocol from Wu et al.²⁰. Mouse brains were fixed in 4% paraformaldehyde (24 hours, 4 °C), cryoprotected in 30% sucrose (3 days) and sectioned coronally at 30-μm thickness using a cryostat. Slices were stored at −20 °C in cryoprotectant. For each group, three sections from six mice per cohort were stained with technical replicates. Sections were rinsed in TBS (3×, 5 minutes), blocked in 2% BSA and incubated overnight at 4 °C with anti-GFP antibody (Aves Labs, cat. no. GFP-1020, 1:250). After washing, samples were incubated with Alexa Fluor 488-conjugated donkey anti-chicken secondary antibody (Jackson ImmunoResearch, cat. no. 703-545-155, 1:250) for 2 hours at room temperature. Excess antibodies were removed with TBS washes. To detect Fe²⁺ and Fe³⁺ simultaneously, DNAzyme-based sensors were used. Active and inactive enzyme (E/iE) strands (2 μM) and substrate strands (2.2 μM) were annealed in Bis-Tris buffer (5 mM Bis-Tris (pH 6), 40 mM sodium acetate, 200 mM NaCl) by heating to 95 °C and cooling to room temperature. Fe(II)-H5 and Fe(III)-B12 sensors were combined 1:1 with Hoechst. Brain slices were rinsed in Bis-Tris-acetate buffer, incubated in the sensor mix for 30 minutes and then washed and treated with 0.5× TrueBlack Lipofuscin Quencher (Biotium, PSF006) in 70% ethanol for 30 seconds. Slides were mounted with Fluoromount-G (SouthernBiotech) and imaged using a Nikon spinning disk or Zeiss LSM 710 confocal microscope with a ×20 objective. Channels used included 405 nm (Hoechst), 488 nm (GFP), 546 nm (Fe²⁺) and 647 nm (Fe³⁺). z stacks (four steps, 5-μm intervals) covering the hippocampus were acquired via tiled scan. Images were analyzed in ImageJ; maximum intensity projections were used. Background correction was done by subtracting the average signal from inactive sensor (iErS) controls in the same region (CA1, CA2/CA3 or dentate gyrus).

DNAzyme sequences

Fe(II)-H5

• aE:/5IAbRQ/TGGATATCTCCTAGCCAGACTGTTATGTGTGATACGGCAAACTTCGTGATGCCTCTACGGGTCCG

• iE:/5IAbRQ/TGGATATCTCCTAGTCAGACTGTTATGTGTGATACGGCAAACTTCGTGATGCCTCTACGGGTCCG-3′

• rS:/5IAbRQ/CGGACCCGTATCAATCTCACGTATrAGGATATCCA/3AlexF546N/

Fe(III)-B12

• E: GCGGCATGCGCGTTTGCGGCACCTAAACGCTCCTAATAGAG/3IAbRQSp/

• iE:GCGGCATGCGCGTTTGCGGCACCTAAACGCCCCCTAATAGAG/3IAbRQSp/

• rS:/5Alex647N/CTCTATTArGGGAGACTCGCATGCCGC/3IAbRQSp/

Electrophysiology

Acute hippocampal slices were prepared from 16–18-week-old male mice, 1 month after stereotaxic injection of Ftl1-overexpressing (Ftl1 OE) or control virus. All procedures were approved by the IACUC of AfaSci (protocol no. 0223). Mice (28–42 g) were deeply anesthetized with halothane and decapitated. Brains were rapidly removed and placed in ice-cold, oxygenated artificial cerebrospinal fluid (ACSF; continuously bubbled with 95% O₂/5% CO₂). ACSF composition (in mM) was as follows: 130 NaCl, 2.5 KCl, 1.2 KH₂PO₄, 2.4 CaCl₂, 1.3 MgSO₄, 26 NaHCO₃ and 10 glucose (pH 7.4). Transverse hippocampal slices (400 μm) were prepared using a tissue slicer (Stoelting) and recovered for ≥1 hour at room temperature in oxygenated ACSF prior to recording. Field excitatory postsynaptic potentials (fEPSPs) were recorded in submerged slices perfused with oxygenated ACSF (flow rate: 1.75 ml min⁻¹; DynaMAX pump) at room temperature. Glass microelectrodes (1–3 MΩ, ACSF-filled) were positioned in the stratum radiatum of CA1 to record responses evoked by stimulation of Schaffer collaterals via a concentric bipolar electrode (FHC, Inc.), placed approximately 100 μm apart. Stimuli were 0.2-ms biphasic pulses (0.4-ms total duration) delivered every 10 seconds. Input/output (I–O) curves were generated using 23 increasing current steps (20-µA increments) from threshold (20–100 µA). Test pulse intensity was set to 30–50% of the maximal response. LTP was induced using two trains of 100-Hz stimulation (100 pulses, 1-second duration, 5 seconds apart). fEPSP slope was measured from the initial negative phase using Clampfit 10.4 (Molecular Devices). Data points represent the average of three consecutive sweeps. Paired-pulse ratio (PPR) was calculated using a 50-ms interpulse interval as the ratio of the second to the first EPSP. Traces were low-pass filtered with a Gaussian filter (−3-dB cutoff: 1,260 Hz). Representative LTP traces are averages of six consecutive responses (1 minute, unfiltered). Recordings were completed within 8 hours of dissection. Data were analyzed using Clampfit 10.4, Excel and StatView 5.0. LTP was expressed as percent of baseline; the final 5 minutes were averaged for group comparisons. Results are shown as mean ± s.e.m. Statistical significance was defined as P < 0.05.

Open field

Mice were placed in the center of an open 40-cm × 40-cm square chamber (Kinder Scientific) with no cues or stimuli and allowed to move freely for 10 minutes. Infrared photobeam breaks were recorded and movement metrics were analyzed by Motor Monitor software (Kinder Scientific).

NOR

The NOR task was performed following White et al.⁵⁴. Mice were habituated to an empty arena for 10 minutes (day 1) and then allowed to explore two identical objects for 5 minutes (day 2). On day 3, one object was replaced with a novel one, and exploration time over 5 minutes was recorded using Smart Video Tracking Software (Panlab, Harvard Apparatus). To control for object and spatial bias, object identity and novel object location were counterbalanced across animals. Objects were selected to elicit exploration regardless of genotype or age. The preference index was calculated as follows: (Time_novel / (Time_novel + Time_familiar)) × 100, where 100% indicates full novel object preference. Mice failing to explore both objects during training were excluded.

Y maze

Spatial recognition was evaluated in a Y maze as described previously. During training, mice explored the start and trained arms for 5 minutes, with the third (novel) arm blocked. Maze arms were alternated and cleaned between trials. After a 45-minute delay, mice were reintroduced and allowed to explore all three arms for 5 minutes. Entries and time in each arm were tracked (Smart Video Tracking Software). Percent arm entries were computed from the first minute of exploration. The discrimination index was calculated as follows: (Time_novel − Time_trained) / (Time_novel + Time_trained). Mice making fewer than three entries in the first minute were excluded.

Radial arm water maze

Spatial learning and memory was tested using an eight-arm radial arm water maze (RAWM) protocol⁵⁵. Mice were trained to locate a constant goal arm, with the start arm varied between trials. Each incorrect entry was scored as an error. On day 1, mice underwent 12 training trials (blocks 1–4; alternating visible/hidden platforms), followed by three hidden-platform trials after a 1-hour break (block 5). On day 2, animals completed 15 hidden-platform trials (blocks 6–10). Errors were averaged per three-trial block. Experimenters were blinded to treatment and genotype during scoring.

Mice NADH treatment

NADH (300 mg kg⁻¹; Roche, cat. no. 10128023001) as well as vehicle (sodium chloride solution, Sigma-Aldrich, cat. no. S8776) were administrated intraperitoneally with an injection volume of 200 µl per mouse. All animals were treated with vehicle or NADH once daily for 9 days and for 2 hours pretrial.

Primary cell culture NADH treatment

NADH (200 uM; Roche, cat. no. 10128023001) as well as vehicle (sodium chloride solution, Sigma-Aldrich, cat. no. S8776) were administrated for five consecutive days starting DIV14 until DIV18 included.

Data, statistical analyses and reproducibility

All experiments were randomized and blinded by an independent researcher. Researchers remained blinded throughout histological, biochemical and behavioral assessments. Groups were unblinded at the end of each experiment on statistical analysis. Data are expressed as mean ± s.e.m. The distribution of data in each set of experiments was tested for normality using the D’Agostino–Pearson omnibus test or the Shapiro–Wilk test. Statistical analysis was performed using GraphPad Prism version 8.0, version 9.0 or version 10 (GraphPad Software). Means between two groups were compared using two-tailed unpaired Student’s t-tests. Additional statistical details are indicated in the respective figure legends. All data generated or analyzed in this study are included in this article. The main experimental findings are representative of two independently performed experiments. All replication attempts were successful. RNA-seq and proteomics data were not replicated due to resource limitations but were orthogonally validated.

Experimental replication was not attempted for negative data.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.